This repository contains tools for downstreaming OLM software
OLMv0
The OLMv0 code is downstreamed into the openshift/operator-framework-olm repository from the following upstream repositories:
The operator-lifecycle-manager
repository is the main repository, as it refers to the others. Its go.mod
file determines what versions of api
and operator-registry
are used and downstreamed. Any integration between these three repositories must be done upstream, as that will provide a clean downstreaming experience.
The upstream repositories must also be go mod tidy
clean.
The downstream repository is a "monorepo" consisting of all the above repositories each located in a staging
directory. The upstream commits are cherry-picked with updating comments indicating the source repository and commit SHA.
In the monorepo there are individual Dockerfiles controlling downstream builds for each tool. For instance: https://github.com/openshift/operator-framework-olm/blob/master/operator-lifecycle-manager.Dockerfile
OLMv1
The OLMv1 code is downstreamed into the three separate repositories:
The upstream commits are downstreamed as a direct mirror (i.e. commit SHAs remain the same) and then merged into the main
branch.
However operator-controller
has dependencies on the catalogd
repo, and requires that its go.mod
is updated to reflect the downstream (openshift) repositories.
- Use the upstream
go.mod
file in operator-controller
to determine which commits of catalogd
are to be merged.
- Synchronize all three upstream repo to their respective downstream repos.
- If there are no changes to
catalogd
(i.e. they've been merged), update the downstream go.mod
file in operator-controller
to reference those downstream repos.
Only if there are no outstanding merges for downstream catalogd
, is the go.mod
file in downstream operator-controller
updated.
Merging to downstream consists of:
- Merging via merge commit upstream
main
branch, this overrides the existing main
branch via git merge --stategy=ours
. This keeps the upstream and downstream commits numbered with the same SHA.
- Then cherry-pick commits as needed:
- Dropping commits that have the headline:
UPSTREAM: <drop>:
- Keeping commits that have the headline:
UPSTREAM: <carry>:
- Determine if a cherry-pick needs to be kept
UPSTREAM: 1234:
by looking at what's being merged.
- Finally, do any post
go mod
and make manifests
-type processing.
Building
Just run make
or make build
. Two executables will be created in the root of the repository
v0
- for downstreaming OLMv0
v1
- for downstreaming OLMv1
Periodic Jobs
This tool is intended to be run as a periodic job with minimal human interaction.
The jobs are defined in:
Manual Merging
Running the merge tools manually will allow you to do the merging in a local repository to fix any issues that the tool itself cannot handle.
Running the tool without specifying the -mode
option will just list the commits that need to be merged.
Running the tool with the -mode=synchronize
will perform the actual merge in the local repository. A PR can then be generated from the results.
Running the tool with the -mode=publish
will perform the merge in the local repository, and then attempt to publish the PR. However, if you don't have the correct credentials set, it will still print the contents of the PR description that can be used to manually create a PR.
See the examples directory for a set of sample scripts.
Cleanup
After merging, the tools may leave cruft behind in the local repository that may interfere with subsequent runs of the tool and/or other git operations.
git clean -fdx
git reflog expire --expire-unreachable=now --all
git gc --prune=now
git worktree prune
The periodic jobs are run in containers, and don't need any cleanup. You may also want to consider running the tool in a clean set of repositories.
Options/Flags
Options and flags may be found in the code: